Google’s “Funny” Algorithm Can Measure Laughter. Well, Sort Of.
By on February 13th, 2012

YouTube has an interesting comedy slam page where users can compare two videos and vote the one which appeals to them. YouTube slam, which is categorized into music, comedy, bizarre and cute, is sort of a never ending addictive game where you watch bizarre videos and vote the ones you find interesting, funny or humorous. You get nothing out of this, except a mention as a top voter on the slam leaderboards.

But have you ever wondered how Google measures which video is funnier, which one is bizarre and which one is cute? What are the signals Google counts when ranking videos in the YouTube slam page?

youtube-comedy-slam

Unlike textual content, Google and other search engines can’t “read” the content of a video or see through it. The only way a bot can have some clue about a video is its associated metadata, the title tags, description, anchor text, links from other domains, user comments and so forth. This mechanism is often, not so accurate and can be quirked for selfish reasons.

Coming to how YouTube ranks videos in slam pages, a Google researcher admitted that there is an algorithm which tries to measure the degree of laughter associated with a video. The algorithm is not entirely driven by view counts or mere votes because of the fact that human feelings are subjective in nature. What appears humorous to one age group might be a complete turn off for another.

youtube-commentsThe funny thing is that the algorithm tries to extract “textual laughter” from the comments section of a YouTube video. Words such as “hahaha”, “hehehe”, “jajaja”, “kekeke” in user comments counts as a strong signal towards the humorous nature of a video. This algorithm also considers web acronyms such as LOL, ROFL or LMAO and emoticons e.g :D, ;-) xP.

Google trains classifiers to identify funny videos and finds the reason why this video is funny and to what category or genre this video may belong to. The ranking algorithm is strongly driven by human reaction and their emphasis on emotions e.g a looooooool  is considered a stronger signal than a lol because of the elongation which reflects the viewer’s emphasis. Here is what Google says:

We noticed that viewers emphasize their reaction to funny videos in several ways: e.g. capitalization (LOL), elongation (loooooool), repetition (lolololol), exclamation (lolllll!!!!!), and combinations thereof. If a user uses an “loooooool” vs an “loool”, does it mean they were more amused? We designed features to quantify the degree of emphasis on words associated with amusement in viewer comments. We then trained a passive-aggressive ranking algorithm using human-annotated pairwise ground truth and a combination of text and audiovisual features. Similar to Music Slam, we used this ranker to populate candidates for human voting for our Comedy Slam.

Now they are judging candidates for YouTube slam by how many “lols” are written underneath a video? Like this video which got more than 3000 comments and 1,566,000 views but it’s not funny!

Tags: ,
Author: Amit Banerjee Google Profile for Amit Banerjee
Amit has been writing for Techie Buzz since early 2009 and keeps a close eye on web apps, Google and all things Tech. He also writes at his own tech blog, Ampercent. Follow him on Twitter @ amit_banerjee

Amit Banerjee has written and can be contacted at amit@techie-buzz.com.
 
Copyright 2006-2012 Techie Buzz. All Rights Reserved. Our content may not be reproduced on other websites. Content Delivery by MaxCDN